[core][imageregistration] Modify shift detection to work with higher values of shift

darfix/core/imageRegistration.py

@@ -26,7 +26,7 @@
 
 __authors__ = ["J. Garriga"]
 __license__ = "MIT"
-__date__ = "26/05/2021"
+__date__ = "16/06/2021"
 
 
 import enum
@@ -169,7 +169,7 @@ def normalize(x):
     return x / numpy.linalg.norm(x)
 
 
-def improve_linear_shift(data, v, h_max, h_step, nimages=None, shift_approach="linear"):
+def improve_linear_shift(data, v, h, epsilon, steps, nimages=None, shift_approach="linear"):
     """
     Function to find the best shift between the images. It loops ``h_max * h_step`` times,
     applying a different shift each, and trying to find the one that has the best result.
@@ -195,23 +195,22 @@ def improve_linear_shift(data, v, h_max, h_step, nimages=None, shift_approach="l
         iData = numpy.random.choice(iData, nimages, False)
 
     score = {}
-    utils.advancement_display(0, h_max, "Finding shift")
-    for h in numpy.arange(0, h_max, h_step):
+    utils.advancement_display(0, 2 * epsilon, "Finding shift")
+    for h_ in numpy.arange(h - epsilon, h + epsilon, 2 * epsilon / steps):
         result = numpy.zeros(data[0].shape)
         for iFrame in iData:
-            shift = h * v * iFrame
+            shift = h_ * v * iFrame
             result += apply_shift(data[iFrame], shift, shift_approach)
 
         # Compute score using normalized variance
         # TODO: add more autofocus options
-        score[h] = normalized_variance(result)
-        utils.advancement_display(h + h_step, h_max, "Finding shift")
-
+        score[h_] = normalized_variance(result)
+        utils.advancement_display(h_ - epsilon, 2 * epsilon, "Finding shift")
     optimal_h = max(score.keys(), key=(lambda k: score[k]))
     return optimal_h
 
 
-def shift_detection(data, h_max=0.5, h_step=0.01):
+def shift_detection(data, h_max=0.5, steps=1000, shift_approach="linear"):
     """
     Finds the linear shift from a set of images.
 
@@ -232,7 +231,10 @@ def shift_detection(data, h_max=0.5, h_step=0.01):
             second_sum += data[i]
     shift = find_shift(first_sum, second_sum, 1000)
     v = normalize(shift)
-    h = improve_linear_shift(data, v, h_max, h_step)
+    v_ = 2 * shift / len(data)
+    h = numpy.sqrt(v_[0]**2 + v_[1]**2)
+    epsilon = 0.5 * h
+    h = improve_linear_shift(data, v, h, epsilon, steps, shift_approach=shift_approach)
     return numpy.outer(h * v, numpy.arange(len(data)))